Well, in the case of most robots, we won't just want to take a single image and classify the dominant image in it. We'll want to detect multiple objects, in each image, and we'll likely be dealing with video. So we can do this by using the TensorFlow Object Detection API https://github.com/tensorflow/models/tree/master/research/object_detection on each video frame! We for sure won't be getting this to work on 60 frames per second using a Raspberry Pi, but if we got even 5 fps that would be pretty cool!